> EG - Take the same inputs -> eye tracking / natural language feedback (speech!) / interface observation / time to goal / etc <- and then let a flush ML / AI c

> EG - Take the same inputs -> eye tracking / natural language feedback (speech!) / interface observation / time to goal / etc <- and then let a flush ML / AI come up with a set of experiments and pathway. I like that... but that's going to take a lot of work to keep it from becoming the contemporary equivalent of "microsoft clippy" but from Hell. Has anyone pursued or published about such an approach yet?